Data analyses were performed using IBM SPSS (IBM Corp. Released 2020. IBM SPSS Statistics, version 27.0. Armonk, NY: IBM Corp) and GraphPad Prism (version 9.0.0, GraphPad Software, San Diego, CA, USA).
First, the total sample was stratified to create two extreme groups, in a similar manner as previous PiL investigations (i.e., [
23,
32‐
34]). With this aim, we used the “visual binning” function from the SPSS, which divides all the included subjects according to a specified number of cut points. The option “equal percentiles based on scanned cases” was used to create 5 sub-groups: Q1 (PiL values: ≤ 21;
N = 146), Q2 (PiL values: 22–25,
N = 135), Q3 (PiL values: 26–27,
N = 114), Q4 (PiL values: 28–30,
N = 129), and Q5 (PiL values: ≥ 31,
N = 100). We later conducted analyses focusing on the extreme groups: the Q1, named lower PiL (LP) group, and the Q5, named higher PiL (HP) group. These two groups shaped our sample of interest, conformed by 246 individuals. Basic demographic data (age, gender, YoE) was directly compared between the PiL groups through a one-way analysis of variance (ANOVA) and a chi-squared test. Cognitive data was integrated into three composite scores: an episodic memory composite (EMc), an executive functioning composite (EFc), and a working memory composite (WMc). The EMc was calculated considering the three recall measures from the Rey Auditory-Verbal Learning Test (immediate, delayed, and recognition). The EFc measure was computed considering the Reasoning Matrix, Block Design Test, Digit-Symbol Substitution Test and Cancellation subtests from WAIS-IV. The WMc measure was calculated with the inverse Digit Span and the Letter-Number Sequencing. Composites were obtained through factorial analyses with SPSS. Brain burden was evaluated considering the total estimation of WMLs. rs-fMRI analyses were computed using SyS data on the 14 Shirer circuits [
59]. To investigate the neuropsychological differences between the PiL groups, a multivariate general linear model (GLM) was conducted considering all cognitive measures together as dependent variables. Moreover, a univariate GLM with WMLs as the dependent variable was calculated to study the brain burden group differences. Subsequently, Pearson correlations between cognitive status and WMLs in each group, as well as slope differences between the groups, were computed. This latter analysis was conducted with regression functions from GraphPad Prism. In addition, to investigate rs-fMRI differences, a multivariate GLM was undertaken considering the 14 Shirer networks altogether. Whether significant group differences emerged on a whole functional system, a subordinate zoom-in was conducted focused on its ROI-to-ROI functional couplings through multiple GLM analyses, considering within- and between-brain network connectivity. As per its exploratory nature, these analyses were not corrected for multiple comparisons. Finally, Pearson correlations were calculated to relate rs-fMRI measures with cognitive performance in each group. In all the stated statistical analyses, age and gender were used as covariates. YoE were included as a covariate when cognitive data was examined. Moreover, all rs-fMRI explorations were also controlled for FWD. All statistical analyses were two-tailed, and
α was set at 0.05. For the Pearson correlation analyses, a bootstrapping with 5000 samples was also applied, and the bias-corrected and accelerated 95% confidence interval (CI) was reported. Quality checks were conducted using the stated covariates to corroborate whether the main associations were also present in the whole sample, as well as to further investigate the role of specific variables (i.e., age). Data in plots are presented with standardized
Z scores, considering the main variables (cognition, WMLs, rs-fMRI) as well as the covariates included in each model (age, gender, YoE, FWD).